Claude Code can run a whole org chart of Claudes. Put a heavy-thinking model like Opus (or Fable) at the top and give it exactly two jobs: hand out assignments and verify that each one comes back complete. This allows it to act like a manager that stays out of the work (and keeps the work out of its context window).
Below the orchestrator, use teams of two. One subagent to complete the task, and a second to check the result. You can run as many teams in parallel as the job needs. When the teams finish, the orchestrator hands one last reviewer nothing but the original plan. This helps provide a more impartial review and provides multiple levels of review, helping ensure higher quality when the task returns to you.
The /tasks command lists what's running in the background. Anthropic now ships an official version of this pattern (agent teams), but it's experimental and burns more tokens than subagents. Clue #12 spawned the Claudes and Clue #9 made them argue. This one gives them a manager.
Roscoe Rawson built this org chart before Anthropic shipped one; the post itself was drafted by a single Claude with no team under it, which felt a little lazy.